Cough detection using a non-contact microphone: A nocturnal cough study

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Nocturnal cough detection is challenging due to confounding sounds like snoring.
• Existing methods may require uncomfortable contact sensors.
• Accurate monitoring of cough frequency is crucial for respiratory health assessment.

Purpose of the Study:

• To develop and evaluate an automatic, non-contact cough detection system for nighttime audio recordings.
• To compare the performance of a Deep Neural Network (DNN) against a Gaussian Mixture Model (GMM) for cough detection.
• To analyze cough event frequency across different sleep stages and its correlation with Obstructive Sleep Apnea (OSA) severity, Body Mass Index (BMI), and demographics.

Main Methods:

• Utilized a database of nocturnal audio signals from 89 subjects recorded during polysomnography.
• Implemented and tested two classifiers: Gaussian Mixture Model (GMM) and Deep Neural Network (DNN).
• Analyzed detected cough events in relation to sleep stages, OSA severity, BMI, and gender.

Main Results:

• The DNN-based system achieved 99.8% accuracy, with 86.1% sensitivity and 99.9% specificity.
• Cough events were significantly more frequent during wakefulness and less frequent during deep sleep.
• Positive correlations were observed between BMI and nocturnal coughs, and between coughs and OSA severity in men.

Conclusions:

• The DNN-based non-contact cough detector demonstrates high accuracy and effectiveness for nighttime audio recordings.
• The system can reliably distinguish coughs from other sleep-related sounds.
• This technology offers a valuable tool for monitoring respiratory illnesses and treatment responses non-invasively during sleep.